Oncofusion

start / graph / ana-the-spike-in-batch-is-in-none-of-the-four-subseries-and-the-ratio-is-in-no-table

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.

Hunting the one reagent that would unbind every share-based result in this thread: the fly/human ratio is in no supplementary table and the words dm6 and scale factor appear zero times in the full text - and the 181116 batch the processing line names is in NONE of the 78 samples across all four subseries of GSE127508, so no deposited sample is established to carry the spike-in at all

also called hunting the discarded spike-in · the 181116 batch is nowhere · no deposited sample is established to carry spike-in · the scale factor is recomputable for 23 GB

Every measurement in this line of work is a proportion, which means it cannot tell a real global change from a bookkeeping effect. The fix is a "spike-in": a known amount of fruit-fly material added to each sample as a ruler. One key dataset's notes say a fly spike-in was added and then judged unnecessary. Chasing it down shows the number was never published, and - more awkwardly - the specific batch of samples the note refers to is not in the dataset at all. It can still be recovered by re-reading the raw sequencing files, and this works out what that would cost.

The same thing, for a scientist

**Statement.** Two beats in one day ended on the same sentence: *the missing reagent is a spike-in, not a better statistic.* The corpus already knew one existed - `ana-q118-gse127507-ran-a-drosophila-spike-in-measured-it-and-deliberately-did-not-use-it` records the verbatim GEO line and filed two open questions. **Neither was ever run. Both are run here, and the second changes the picture.** **`S2` — THE RATIO IS NOT IN THE PAPER.** Full text of **PMC7227117** (PMID **31821784**, Krivtsov AV et al., *Cancer Cell* 2019, DOI 10.1016/j.ccell.2019.11.001): **`dm6` 0 hits, `scale factor` 0 hits.** Of three `spike` hits, two are VTP-50469 *spiked-in chow* and one is the protocol. **Six supplementary captions, none about spike-in or normalisation** — S1 gene expression, S2 occupancy, S3 PDX samples, S4 occupancy loss, S5 GSEA gene sets. *"Small"* stays unquantified. **`B2` — AND THE BATCH IS NOWHERE.** The processing line names the **181116** samples. Across **all 78 samples of all four subseries** — GSE127506 (42), GSE127507 (26), GSE137813 (6), GSE137814 (4) — **the stamp appears in ZERO titles.** It occurs 52 times, all inside GSE127507's shared boilerplate, twice per sample. The stamps that do appear are 180927 ×36, then 180110 ×10, 180402 ×4, 180619 ×3, 180726 ×2, 180919 ×2, 181213 ×2, 190211 ×2, and 190710 ×4. **So the corpus's reading was too generous to the deposit.** It is not that GSE127507's samples carry an unused spike-in. **It is that the sentence describes a batch that is not in this SuperSeries**, and no deposited sample is established to contain fly chromatin at all. **One new fact cuts the other way: the protocol dose IS published — `10pM` of *Drosophila melanogaster* chromatin**, added at the EDTA/EGTA stop of a Protein-A-MNase digestion. If any of these libraries did receive it, **the fly reads are physically in the raw data.** **`C2` — RECOMPUTATION IS A BOUNDED JOB.** NCBI exposes no precomputed species breakdown (`run_taxonomy` returns 0 bytes, the analysis endpoint returns *does not exist*). GSE127507's 26 experiments resolve to **26 paired NextSeq runs, SRR-level, 848.2M spots, 23.2 GB, mean 915 MB, largest 1,952 MB**. A full realignment is cluster work — **but the fly fraction is a PROPORTION**, so ~10⁶ reads per run estimates it to well under a tenth of a per cent. **23 GB becomes a small job.** **And the same subsample answers the harder question.** Counting fly reads tells you directly whether these libraries contain spike-in DNA *at all* — which the metadata cannot, because the boilerplate is ambiguous between *"a batch that is not here had spike-in"* and *"all these samples did and the label was never used in titles"*. **SUPERSEDED IN PART, TWENTY MINUTES LATER, IN THIS SAME CYCLE.** This entry concluded that recovering a scale factor would mean recomputing from **23.2 GB of SRA**. That is true of GSE127507 and it was the wrong question. **A GEO search for the same lab's spike-in work returns `GSE149183`, whose processing line is the SAME boilerplate with the OPPOSITE outcome** - *"Global differences in fly-human chromatin ratios WERE FOUND, so H3K79me2 TSS signals were normalized accordingly"* - **and whose series-level supplementary files contain the dm6 counts and per-sample ratios in 2 MB of plain text.** One of the two is **RS4;11, a KMT2A::AFF1 line - this project's index fusion.** **The negative above stands for GSE127507. The conclusion drawn from it - that the ruler is expensive - does not.** See `ana-the-spike-in-ruler-is-deposited-in-plain-text-and-the-global-change-is-5-to-23-fold`.

Go deeper
What the metadata cannot decide, and why that matters

The processing block is identical across all 26 GSE127507 samples and contains two sentences that sit badly together: one names the 181116 samples as carrying spike-in, the other says these samples were additionally aligned to dm6 — and dm6 occurs exactly 26 times, once per sample.

Two readings survive the metadata. Either the block is boilerplate inherited from a batch deposited elsewhere, and these 26 samples have no fly chromatin; or every one of them does and the batch label was simply never propagated into titles. Nothing in GEO distinguishes them, and the difference decides whether a scale factor exists for this series at all.

This is why C2 matters more than S2. A literature search can only ever establish that a number was not printed. Counting reads establishes whether the molecule was in the tube.

A regex boundary that reported the wrong answer, and how it was caught

The first pass delimited date stamps with \b and reported zero stamps in every subseries — including in titles like ML2_ChIP_DOT1L_DMSO_180726_TGGTTGTT. \b cannot match between _ and a digit, because both are word characters.

It was caught because the q118 entry had already published the stamp distribution for GSE127507, and zero contradicted it. The corrected pattern uses explicit digit lookaround and reproduces that published distribution exactly — 180110 ×10, 180209, 180402 ×4, 180619 ×3, 180726 ×2, 180919 ×2, 181213 ×2, 190211 ×2. An earlier entry's published number is the cheapest available test of a new instrument, and it is worth writing one down for that reason alone.

Facts and figures
questionis the discarded Drosophila spike-in recoverable, and where is the 181116 batch?
answers open questions fromana-q118-gse127507-ran-a-drosophila-spike-in-measured-it-and-deliberately-did-not-use-it
supplementary captions6
any about spike infalse
protocol dose published10pM Drosophila melanogaster chromatin, added at the EDTA/EGTA stop of a Protein-A-MNase digestion
total samples checked78
titles containing 1811160
the undecidable bitthe shared processing block is ambiguous between 'a batch not deposited here had spike-in' and 'all 26 did and the label never reached the titles'; GEO cannot separate them and counting fly reads can
absence classmeasured-not-curated
bite backThis entry argues for spending compute on recovering a scale factor. If the libraries turn out to contain no fly DNA, that compute buys a negative - and the entry will have sent someone after a reagent that the metadata already hinted was not in the tube.
full text hits
dm60.000dm6: 0scale_factor0.000scale_factor: 0Drosophila1.000Drosophila: 1spike3.000spike: 3
subseries checked
GSE1378144.000GSE137814: 4GSE1378136.000GSE137813: 6GSE12750726.0GSE127507: 26GSE12750642.0GSE127506: 42
occurrences of 181116
GSE1275060.000GSE127506: 0GSE1378130.000GSE137813: 0GSE1378140.000GSE137814: 0GSE12750752.0GSE127507: 52
paper
{
  "pmid": "31821784",
  "pmc": "PMC7227117",
  "doi": "10.1016/j.ccell.2019.11.001",
  "journal": "Cancer Cell",
  "pubdate": "2019 Dec 9"
}
full text hits — the numbers behind the chart
{
  "dm6": 0,
  "scale_factor": 0,
  "spike": 3,
  "Drosophila": 1
}
subseries checked — the numbers behind the chart
{
  "GSE127506": 42,
  "GSE127507": 26,
  "GSE137813": 6,
  "GSE137814": 4
}
occurrences of 181116 — the numbers behind the chart
{
  "GSE127507": 52,
  "GSE127506": 0,
  "GSE137813": 0,
  "GSE137814": 0
}
sra
{
  "experiments": 26,
  "runs": 26,
  "total_GB": 23.2,
  "spots_millions": 848.2,
  "mean_run_MB": 915,
  "largest_run_MB": 1952,
  "layout": "PAIRED",
  "platform": "ILLUMINA NextSeq 500",
  "precomputed_taxonomy_available": false
}
verdicts
{
  "S2": "the ratio is in no table caption and no text",
  "B2": "the named batch is in none of the 78 samples",
  "C2": "recomputation is a bounded job; the fly FRACTION needs only a subsample"
}
SUPERSEDED IN PART
{
  "what": "the cost conclusion. The ratio really is absent from GSE127507's paper and the 181116 batch really is in none of the 78 samples - but a sibling deposit, GSE149183, has deposited dm6 counts and ratios as plain text",
  "found_by": "a GEO search on the submitting author plus the term spike-in, one query",
  "why_it_matters": "this entry reasoned from one deposit to a cost estimate for the whole question. The lesson is the standing one: before publishing a negative, search for the better reagent",
  "see": "ana-the-spike-in-ruler-is-deposited-in-plain-text-and-the-global-change-is-5-to-23-fold"
}
Still open
  1. SUPERSEDED IN PART: GSE149183 deposits dm6 counts and ratios in plain text, including for RS4;11 (KMT2A::AFF1). The 1e6-read subsample experiment below is still the only way to settle whether GSE127507's own libraries contain fly DNA, but it is no longer the cheapest route to a scale factor.
  2. Subsample ~1e6 reads from each of the 26 runs and count reads assigning to dm6. That single experiment decides whether a scale factor exists for GSE127507 AND whether these libraries contain fly chromatin at all - two questions no amount of metadata reading can separate.
  3. Is the 181116 batch deposited under a DIFFERENT accession entirely? It is in none of GSE127508's four subseries; a GEO search on the stamp, or on the same submitter, would find it if it exists.
  4. Even a recovered scale factor applies to GSE127507. Is it transferable to GSE260742 or GSE299100, where the results that need it were computed? Probably not, and that should be argued explicitly rather than assumed either way.
  5. Ask the same question of GSE299100, which has never been checked for a spike-in at all.
Every source behind this entry (1)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis

    The full text of PMC7227117 contains zero occurrences of 'dm6' and zero of 'scale factor'; of three occurrences of 'spike', two refer to VTP-50469 spiked-in chow and one is the methods sentence 'spike-in 10pM of Drosophila melanogaster chromatin'. Six supplementary-material captions are present and none concerns spike-in or normalisation. Across all 78 samples of GSE127506, GSE127507, GSE137813 and GSE137814, the string 181116 appears in zero sample titles and 52 times in GSE127507's shared processing lines, twice per sample; 'spike' appears 52 times and 'dm6' 26 times, both only in GSE127507. GSE127507's 26 experiments SRX5444508-SRX5444533 resolve to 26 paired-end Illumina NextSeq 500 runs totalling 848.2 million spots and 23.2 GB, mean 915 MB and largest 1952 MB. The NCBI run_taxonomy endpoint returns zero bytes for SRR8646485 and the analysis endpoint returns an error.

    What this evidence cannot show

    THE PMC XML CARRIES TABLE CAPTIONS, NOT SPREADSHEET CONTENTS. S2 means the ratio is not in the text or in any caption; a number inside an .xlsx would be invisible to this test and S2 must never be read as 'not published'. A date stamp is a naming convention, so B2 bounds what the titles say and not what happened at the bench - samples from the 181116 batch could have been retitled, and the shared processing block is ambiguous between two readings that the metadata cannot separate. The SRA footprint is what NCBI reports, not a measured download. The claim that 10^6 reads suffices is a binomial argument about estimating a proportion and assumes fly reads are distributed across the run rather than clustered by tile or adapter batch.

    Evidence that argues against this (3)
    • B2 is the interesting result and it is the weaker of the two. 'The stamp is in no title' is compatible with every one of these samples containing spike-in - GEO titles are not a manifest, and the depositors may simply have stopped using the batch label. Reading it as 'no deposited sample is established to carry spike-in' is correct but easy to over-hear as 'the samples have no spike-in', which this analysis does NOT show.
    • The cost argument makes recomputation sound cheaper than it is. A 10^6-read subsample bounds the SAMPLING error on the fly fraction, not the systematic error: alignment to dm6 needs a decoy-aware strategy or human reads with fly-like k-mers will be miscalled, and at the small fractions expected that bias can exceed the sampling term. The honest statement is that the sampling cost is small, not that the analysis is easy.
    • Chasing a discarded ratio may be the wrong response to the compositional problem in the first place. Even a perfect scale factor from this series would apply to GSE127507, while the results it is wanted for were computed on GSE260742 and GSE299100. A spike-in in the wrong deposit does not rescue a share in the right one, and this entry does not establish that any of the three is transferable to the others.
Connected to